“It’s a pressing issue because more and more of our business today is moving to these platforms,” he says Robert Bray, associate professor of operations at Kellogg. For platform businesses, “disintermediation is the worst possible thing: it is literally a set of [their] customers training another set of customers to exit the system. Once you teach someone to mediate, you may lose them for good.”
And while it can be hard to build sympathy for the likes of Uber or Postmates, Bray adds that not every platform business is a venture-backed behemoth. He and colleagues Ekaterina Astashkina and Ruslan Momot (of the University of Michigan) and Marat Salikhov (of Russia’s New School of Economics) looked for evidence of disintermediation in an app that connects local cleaners with customers in Moscow. “It does provide a useful service, but they don’t really have that many degrees of freedom to keep people on the network,” he explains. If disintermediation were common, “I don’t think the platform would be sustainable.”
Fortunately for Moscow’s house cleaners—but less so for mainstream economic theory—the researchers found no evidence of disintermediation, a result they called “disturbing.”
“This is an area that has received a lot of theoretical attention,” says Bray. “If you were to model all agents as perfectly rational, it’s pretty clear that everyone has an incentive to mediate as quickly as possible. People say, “this has to happen.” Why can’t we find it?’
Looking for signs
One reason disintermediation is hard to find is that “by design, it’s kind of insidious,” Bray explains. “The people who are doing it are doing it in a way that the platform can’t really see.”
For example, consider an application that facilitates several successful transactions between a service provider and a customer, and then sees both users suddenly leave the platform. Did they conspire to cut out the middleman or did they just have nothing else to do? Usually, there is no way for the platform to make a difference.
But the Moscow-based home cleaning app had a key feature that allowed Bray and his colleagues to tackle this problem: it tracked users’ locations by default, even when they weren’t actively using the app. (This was a common practice in smartphone apps before 2019, when Apple’s iOS 13 began encouraging users to opt out of default tracking.)
These data, which were anonymized by the researchers, provided an unusually transparent way to observe any potential disintermediation. The app recorded the time a cleaner was scheduled to perform a specific task, plus the specific location of the residence where the cleaning took place. But because the app was always by passively tracking the cleaners’ locations, researchers could also see if they returned to clients’ residences without a task formally scheduled through the application. Such visits would be conclusive evidence that disintermediation had occurred.
“Maybe a house cleaner does two or three cleanings for one resident and then the relationship [on the app] it’s just ending. Now the question is: Did they mediate or not?” Bray explains. “We responded to this in the most obvious way – within the next month or so after the suspected termination of the relationship, we looked to see if the cleaner returned to that address. The scene of the crime, as it were.”
The researchers also felt that the app itself was “ripe for disintermediation” for several reasons. First, the platform added significant markups to the wages of domestic cleaners—sometimes as much as 40 percent. Moreover, unlike US clients, Moscow residents at the time had “no cultural precedent” for hiring service workers through official agencies for a commission fee – it was much more common for such transactions to be conducted under the table. (According to the researchers, this shadow economy accounts for more than 40 percent of Russia’s GDP.) Additionally, for anyone wishing to avoid the app’s fees, the personal nature of the house cleaning job would make it easy to arrange future appointments without relying on the platform.
In other words: if ever there was a perfect trap for catching middlemen-cutters in the act, Bray and his colleagues would have found it. “We started looking for disintermediation. We wanted to find it,” he says.
“But we couldn’t.”
A non-smoking gun
What did they find? Not just an absence of evidence for disintermediation—which would be disappointing but expected given how difficult the phenomenon is to observe—but real evidence of absence.
This evidence that mediation it was not What’s happening – think of it as the opposite of a smoking gun – came from comparing several sets of location data from the home cleaning platform.
The first set plotted the probability that a house cleaner would be within 500 meters of a construction site when that job was formally scheduled through the app. In this “work sample,” house cleaners showed up when and where they were expected: the probability that they were within 500 meters of the house they were supposed to clean was about 88 percent. A second set, called the “non-working sample,” measured the opposite effect: among those scheduled for work, there was less than a 3 percent chance they were near a house they had officially cleaned at least once in the past. These two data sets established a clear portrait of “good behavior” for app users.
But the ‘no-smoking weapon’ came from a third set of location data – one showing where cleaners were most likely to go in the months after their final scheduled task, as the application continued to passively track them. This is the time period during which disintermediation is likely to occur. This third data set turned out to look almost exactly the same as the “non-functional sample”: there was only about a 2 percent chance that house cleaners would be within 500 meters of a house they had previously cleaned.
In other words: there was no sneaking around, no cutting out the middleman. If there was, investigators would have seen it in the location data—just as clearly as if it were security footage from a resident’s building.
“It’s probably the first negative result in this,” says Bray. “It’s extremely difficult to estimate platform disintermediation, and all the literature says there should be. We found very strong evidence that it is not.”
Theory versus practice
Bray cautions that no paper can definitively overturn economists’ current understanding of disintermediation. However, he says his team’s results suggest the phenomenon “is a problem in theory, but not in practice” for many platforms.
“Theoretically, the platforms should break up [because] every time you and I trade, we should always agree to screw the platform,” he explains. “But we’ve learned that, actually, the ease of working with the platform is often worth the twenty or thirty percent cut they need. It’s just easier. Therefore, the platforms may not need to take the countermeasures [against disintermediation] this theory would suggest.’
Although their study did not formally investigate why users of the home cleaning app chose not to cut out the middleman, the researchers offered some speculative explanations. First, inviting a stranger into your apartment to clean it “is a very intimate thing, so you might want some security” that a platform provides, Bray says. “Anyone going on the app is already looking for an unmediated relationship.”
As for the platform’s surcharges, “Russia has fairly high income inequality—which means people who are well off enough to use the app can probably afford the few extra rubles.”
As for why his results contradict economic theory, Bray has a hunch about that as well. “There is some psychological bias [in the field] that says positive results are interesting and negative results are not,” he says. This “desk-drawer” bias means that the studies show something lack of to occur—such as disintermediation, in this case—they are less likely to be submitted to journals for publication in the first place.
“I think it’s important sometimes to be willing to post a negative result,” he says. “That’s the only way the truth will come out. We say, “We thought we were all on the same page [about disintermediation], but here’s something that shows we’re not.” No paper is definitive, of course. It just makes the picture a little more complicated.”
